| The challenge for people in information society is to deal with mass data with high abilities. Data minging technique has been risen to the challenge and been a hot research topic. Clustering mining is one of the most important data mining technologies that have been widely used.This paper mainly makes improvement on fuzzy clustering algorithms FCM. A new object function is put forward on the basis of Competitive Agglomeration, Noise Clustering and FRC on relational data. According to the deduce methods mentioned in the above algorithms, the necessary conditions for minimization are derived through use of direct objective function minimization based on the Lagrange multiplier technique. The algorithm has the following advantages: first, as the iteration proceeds, the final partition is taken to have the "optimal" number of clusters based on competitive technique without apriori knowledge by introducing competitive item; second, the algorithm becomes more robust through introducing noise parameter; third, it can be directly used on non-Euclidean data with the dissimilarity parameter. In addition, through doing research on partitioning methods, an improved k-means algorithm is given. Through computer emulation, the advantages of the algorithm are validated.Considering the character of the tunnel damage checking data, effective methods on data preprocessing, similarity computing, clustering and damage evaluation are introduced. The first step is data cleaning, integration and transformation. Second, approaches on similarity computing of tunnel data are introduced. At last, improved fuzzy clustering algorithm is given as well as approach on damage evaluation. Through cluster analysis on 758 tunnel data records, initial cluster results are obtained and divided into 4 ranks according to the damage degree, which correspond to 4 kinds of damage situations. Some suggestions are given in order that they can provide decision-making during the process of prevention and cure.
|
PDF Full Text Request